Claude now speaks Office

Claude, the large language model from Anthropic, has focused most heavily in recent months on becoming a fantastic platform for programmers. It hasn't lagged far behind its main rivals, ChatGPT and Gemini, in performing tasks more typically needed in law, but it hasn't been the class leader for some time. But Claude is back! A new very simple addition to the Claude portfolio of tricks now makes it a very attractive platform for those in law: the ability to work directly with the Microsoft Office ecosystem of products: PowerPoint, Word, and Excel. As I will discuss, it's not a perfect implementation yet. But it's very useful now and likely to become more so if Anthropic can work out a few kinks.

PowerPoint

Claude can now directly create PowerPoints that you can download. Here's an example. I give Claude a Supreme Court decision. In this instance, I provide it with Noem v. Vasquez Perdomo, a recent case involving whether immigration enforcement officials can use a form of profiling that includes "Mexican" ethnicity in determining whom to forcefully investigate for immigration violations. This is my prompt.

Create a Powerpoint presentation of this case for 1L law students. I want 20 slides. You should internally create an outline of the legal and political/practical issues involved, using the web as needed to provide context and understanding. Make sure the slides provide the background so that the students will understand the issues involved. Those issues include the standards for a stay, standing (particularly under the Lyons case), key precedents on immigration stops, nationality/ethnicity discrimination under the equal protection clause, and the nature of irreparable injury. Cover both the Kavanaugh opinion and the Sotomayor dissent, noting we do not have an actual opinion from the court itself. Use your new skills to create an actual Powerpoint presentation. Do not use the Python tool; use the newer tool and skill.

Notice a few things about the prompt. I have done the usual specification of audience (1L law students) and output format (20 slides) that any good prompt should have. I've also coaxed it into doing a little Chain of Thought reasoning: "You should internally create an outline of the legal and political/practical issues involved, using the web as needed to provide context and understanding." I'm not sure that's necessary; modern large language models that have "reasoning" built in sometimes perform better without humans trying to micromanage them. I wanted to make sure, however, that Claude actually augmented its training data with contemporary research before leaping off into an inaccurate analysis. Also note that I have used my expert law professor knowledge and sense of 1L pedagogy to provide Claude some specific guidance as to what the issues are and how I want them presented. Again, I could probably get away with omitting this detail, but it's easy enough to provide and promises to align the presentation with what this human thinks is important. Finally, there is the last sentence borne of a bad experience. Apparently, Claude has some older tools lying around to create PowerPoints. They are lousy. The last sentence seeks to guide Claude to use its new superpowers, which uses JavaScript libraries behind the scenes.

Success! Claude provided a preview and a downloadable link that made putting the PowerPoint on my computer a one-click process. Here's a link so you can review the product for yourself. Here are screen captures of sample slides so you can see for yourself the quality of the result.

Notice that the slides are uncluttered but do a good job in capturing the essence of the opinion, leaving oral presentation from the professor and discussion with the class to fill in details. This approach is probably better than my own inclination (lamented by some students) to convert the PowerPoint into a visual rendition of teaching notes. And please note that there is nothing special about dumping a Supreme Court opinion on Claude. It could create a PowerPoint out of a cluster of documents, your own teaching notes, a YouTube transcript, or essentially anything the AI can ingest. And you don't have to accept a default style of PowerPoint. You can give it guidance. Here, for example, is a slide from a presentation it created in response to the following prompt.

Create a downloadable PowerPoint presentation for law students on this opinion. It should be only 3 slides and have outlandish styling. It should also have an image drawn from the web that captures some of the underlying events involved. Use your most modern tool to create the presentation. Do not use the old python tool.

Clearly, Claude grasped the meaning of "outlandish" quite well.

Word

OK. So it can make nice (or hideous) PowerPoints now. But there's more. It can directly edit Word files and create new ones.

Editing of Word Files

One of the coolest features of new Claude is its ability directly to edit Microsoft Word files. Here, I take a problem for my constitutional law class clearly created at the height of the COVID-19 pandemic.

I could have Claude create an answer to the problem, but it could do that reasonably well before gaining its latest powers. Instead, what I want is a document that shows students how each fact in the hypothetical might be relevant to a constitutional analysis. Here's the prompt:

I want you to produce an edited version of this document in which each fact provided in the hypothetical has an annotation bracketed by <annotation> </annotation> that discusses how it is relevant to the constitutional issue. I want your output to be a one-click download Microsoft Word file.

And look at what Claude does.

It basically teaches students how to relate the facts of a problem to the law. It does so without my having to think (thank goodness) and, even better, without my having to cut and paste from the AI into the Word document. And while I am being somewhat facetious here, this ease of use is the difference between producing a useful document for students and not doing it. Moreover, really I should not be doing this at all for students. Students can, should, and will be doing it for themselves. (They might want to improve on my prompt by adding a request for relevant citations). And, yes, they will have to learn the skills for themselves in case they are stranded on the hypothetical computerless desert island so feared by many of my colleagues, but, still, seeing how the process of legal analysis works with multiple examples is likely to be very useful, particularly when one of those computerless desert islands is called The Bar Exam.

Creation of Word Files

In addition to editing existing Word documents, Claude can create new ones. Let's have it create a one-click download Microsoft Word document that annotates the PowerPoint we produced earlier. Here's the prompt. Note, although I am, as a matter of convenience, using the earlier (non-hideous) PowerPoint as the predicate for the document, there is zero requirement that I do so. Claude should be able to directly place in a Word document whatever response it was previously capable of generating. The new capabilities solve the first world problem of otherwise having to go through multiple clicks in moving from an AI response to a well-formatted document in one of the formats most preferred in the world.

I used this prompt to have Claude go from raw material to a Word document.

Here is a PowerPoint presentation on a very recent Supreme Court order. Create a Microsoft Word document that I can download with one click. The document should be a a script that I can use while showing the PowerPoint or that another AI can use as input in a text-to-speech model. The idea is to have a document that can be narrated live or automatically in some form of asynchronous learning. The script should have about 2500-2800 words altogether. Delineate clearly which comments go with which slide. <slide 1>The Supreme Court case of Noem v. Vasquez Perdomo ... </slide 1> <slide 2> blah blah </slide 2> etc. You should primarily use the Powerpoint itself as the basis for the discussion but you can draw on materials from the web or otherwise as needed for supplementation. The tone should be engaging but under no circumstances should be folksy or condescending. It should be objective though some skepticism as to why strict scrutiny is not required for the nationality/ethnicity component of the ICE standard would be fine. The audience is first year law students. Do not end with "does anyone have questions." Just get right into the material. No throat clearing.

Here is a screenshot of the Microsoft Word file created by Claude that indeed could be downloaded with one click. It took my permission to be skeptical a little far, but otherwise is pretty faithful to what I wanted from the script. The point is not that before September of 2025 Claude would be incapable of generating this sort of material but that Claude now facilitates immediate generation of material in one of the very popular formats created by software giant Microsoft.

Excel

For people in the legal world who may have to use numbers on occasion, Claude now also does Excel. And it does so really well. Here's the prompt I gave it. It's from the world of insurance regulation and requires both data fabrication and some fairly complex actuarial calculations.

Create a multi-sheet excel spreadsheet that I can download with one click. It should contain (1) a realistic actuarial life table with the usual columns, including what is usually denoted as q, and (2) computes the net single premium, net level premium and reserves (given net level premiums) over time on a policy in which the user can input a death benefit, an interest rate, a length of policy, and a mortality risk multiplier that can range from 0.1 to 3. For now, assume no lapses.

The output, produced in less than 5 minutes, is spectacular, one I would venture to say fewer than 1 in 1000 law professors could produce and one that might take even trained actuaries an hour or so to generate. Claude is just really, really good at math and programming. Claude created a three-tab Excel spreadsheet containing the requested life table, a sheet for the user to input key parameters (including the issue age parameter that I forgot to request), and the resulting computations.

Here is the input page:

Here is a screenshot of the resulting computations.

Here's the most difficult computation: the determination of reserves.

I couldn't resist challenging Claude to take on a more challenging problem: add a parameterized lapse rate to the computations. When insured individuals cancel their policies early, they do so before mortality risks significantly increase. In a competitive insurance market, if regulators allow, high lapse rates encourage insurers to reduce premiums. Here's the prompt:

Could you now modify the spreadsheet to permit use of a lapse rate. Let the user set the lapse rate with an initial factor and a decay rate so that the lapse rate in year t is L0 /(1+d)^t where L0 is the initial lapse rate, d is the decay rate and t is the year in question. Assume the insurer operates in a competitive market so it must lower its premiums in response to lapses and there are no regulatory restrictions on it doing so. Assume that lapses are uncorrelated with mortality (obviously not true, but it keeps the math simpler).

Again, Claude solved this actuarial problem in a few minutes and provided a one-click Excel file that did the complicated multiple-decrement math. The input screen, not shown here, allows the user to specify an initial lapse rate (default value 8%) and a decay rate (default value 15%). Claude shows the net single premium without lapse and its substantial reduction as a result of lapses.

For the limited subset of my audience that finds actuarial math almost as interesting as AI, I then challenged Claude to permit the user to correlate mortality risk (inversely) with lapse rates. Claude warned me that to do this well required use of copulas for which Excel provides no direct support. It then advised, however that it could do a crude approximation using correlation coefficients. The resulting spreadsheet is here for the many readers who will want to examine it. I do not vouch for the mathematical correctness of the results.

The Somewhat Bad News

All of this is great. Claude now offers a major convenience: direct integration with the Microsoft Office universe beloved by legal professionals everywhere. And now for the bad news. First, for right now, this feature is available only on paid plans. The free tier does not provide one-click download. But those with limited budgets should not fear. If Claude can do this, surely Open AI and Google can do so too. I cannot imagine either of them is too far behind. Moreover, in the competitive world of AI features that require a paid subscription one year often become free the next.

The second problem is also likely just a matter of time to go away. There is currently what I would regard as a bug or at least a serious limitation in Claude when producing PowerPoints: it runs out of context easily. I found that all it could create in a single session using the Sonnet model was a 25-slide presentation. Follow up questions and requests were met with an out-of-context refusal message. I could get somewhat better results using the Opus 4.1 model but without paying Anthropic a significant monthly sum, Claude places some significant monthly limits on the use of this outstanding model. So, for now, save Opus 4.1 for the most important and complex matters. I have to believe, however, that Anthropic engineers are busy solving whatever coding misbehavior leads to the current limitation.

One more thing

Don't forget that Claude and PowerPoint are hardly the only way to make a good presentation document. As covered in a prior blog entries (here and here), all the major large language models are capable of producing LaTeX code, which can be edited in outstanding free platforms such as Overleaf, downloaded as PDF presentations, and manipulated from there. Using LaTeX also gives you consider flexibility in editing and reformatting. Still, as many law professors are allergic to LaTeX and love good old PowerPoint, the new abilities of Claude should be very welcome.